# Docs AI accuracy **URL:** \ **Category:** Developers\ **Created:** [July 7, 2025, 8:22pm UTC](https://forum.dfinity.org/t/docs-ai-accuracy/52836 "2025-07-07T20:22:08Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![552020](https://sea1.discourse-cdn.com/flex023/user_avatar/forum.dfinity.org/552020/32/20749_2.png) [@552020](https://forum.dfinity.org/u/552020)\ **Post date:** [July 7, 2025, 8:22pm UTC](https://forum.dfinity.org/t/docs-ai-accuracy/52836/1 "2025-07-07T20:22:08Z")
I tested today how different AIs know the IC docs. I tested both a well-known AI and the ‘official’ Docs AI. Background is that the well-known AI drove me crazy with a `vetkd_ibe_decrypt_key` method that doesn’t exist or doesn’t exist anymore. So I started with some general test about how well the well-known AI knew the Rust CDK. It gave a detailed but pretty **bad** answer. I let the AI of the Docs evaluate that answer, and it confirmed that it was a **good** one, with only some minor issues. I then compared the answer with the `ic-cdk` crate myself ([https://docs.rs/ic-cdk/latest/ic\_cdk/bitcoin\_canister/index.html](https://docs.rs/ic-cdk/latest/ic_cdk/bitcoin_canister/index.html)) which revealed that the answer was not good at all. After feeding the index page into the well-known AI, I let the well-known AI come up with some questions that could have tricked the Docs AI. The question I picked was: > Does the `ic-cdk` crate expose a `bitcoin_canister` module? The answer of the Docs AI was: > Based on the provided knowledge sources, the `ic-cdk` crate does not currently expose a `bitcoin_canister` module in its stable release. This is confirmed by a forum post where a user could not find such a module in the `ic-cdk` repository, and an official response indicated that this functionality was not yet included in the stable Rust CDK. Instead, developers are advised to invoke the Bitcoin canister directly for such functionality [Forum Post: How to access the bitcoin block headers using the API?](https://forum.dfinity.org/t/34340). > If you need to interact with Bitcoin-related APIs, you may need to construct the calls manually or use other available interfaces until the Rust CDK provides direct support. Which is pretty wrong, considering: [https://docs.rs/ic-cdk/latest/ic\_cdk/bitcoin\_canister/index.html](https://docs.rs/ic-cdk/latest/ic_cdk/bitcoin_canister/index.html) Is the problem known and there is a way to keep the AI of the Docs updated with all the relevant docs? Because it is a pretty useful instrument. I know that RTFM is the answer. But for all the lazy people around could not be possible to let also the AI of the Docs RTFM? ---
**Author:** ![Vivienne](https://sea1.discourse-cdn.com/flex023/user_avatar/forum.dfinity.org/vivienne/32/29170_2.png) [@Vivienne](https://forum.dfinity.org/u/Vivienne)\ **Post date:** [July 8, 2025, 7:47am UTC](https://forum.dfinity.org/t/docs-ai-accuracy/52836/2 "2025-07-08T07:47:58Z")
The Docs AI does RTFM regularly (maybe every 2 weeks?), but it is still an AI. I suppose in this case the problem is that for a long-ish time the way to talk to the BTC canister you were supposed to go through the management canister, so a fair amount of older content plus most of the forum posts (which it also reads) all point in a different direction. Most likely if you pick something that has been around for a little longer in a stable state (0.18 of the CDK was finally stably released 3 weeks ago, the version before does [not contain the bitcoin\_canister module](https://docs.rs/ic-cdk/0.17.2/ic_cdk/?search=bitcoin_canister)) where a little more content is avialable the results would be better. TLDR: yes, we know it’s flawed, but we don’t have anything better at the moment…